Implementing AI Video Quality Enhancement in Mobile Apps

Implementing AI Video Quality Enhancement in Mobile Apps

Clients often come with video files shot on old cameras or compressed to save space: 480p, 1 Mbps bitrate, noise in shadows. The task is to boost to 1080p, remove compression artifacts, stabilize shakes. Photo processing won't work: video is 30 frames per second, each frame must be processed quickly while maintaining consistency between them.

The solution splits into two modes: post-processing (a 1-minute clip in 2–3 minutes) and real-time (33 ms per frame at 30 fps). We have implemented dozens of such pipelines for iOS and Android — from simple upscaling to complex temporal enhancement. With 10+ years of experience, over 50 successful projects, and 5 years on the market, we guarantee quality.

Order AI video enhancement development: we'll select the architecture for your scenario.

How to Choose a Model for a Mobile Device?

The model choice is determined by the target resolution and available hardware. For post-processing we use per-frame models (e.g., Real-ESRGAN) — they are easy to deploy but suffer from temporal flickering.

Temporal models (BasicVSR++ with a 3–5 frame window) deliver smooth results but require batch processing and more memory. Temporal models eliminate flickering, achieving 2x better visual quality than per-frame models. On mobile devices we use lightweight versions with 256×256 tiling. For real-time — lightweight custom models (2–4 MB) designed for 360–480p input and accelerated via Core ML or TFLite GPU Delegate.

Implementation Steps

1. Choose the model type based on target resolution and hardware: per-frame for simple upscaling, temporal for flicker-free results.
2. Set up video decoding with AVAssetReader (iOS) or MediaCodec (Android), ensuring efficient color space conversion.
3. Implement ML inference using Core ML on iOS or TFLite GPU Delegate on Android, optimizing for tiling if needed.
4. Apply temporal post-processing: average activations across 3–5 neighboring frames to eliminate flickering.
5. Encode the processed frames with AVAssetWriter or MediaCodec, and copy the original audio track with PTS synchronization.
6. Test on multiple devices and edge cases, including non-standard resolutions and rotation.

Implementation on iOS and Android

iOS: We use AVAssetReader to decode into CVPixelBuffer, convert YUV→RGB via Metal shader, run through a Core ML model, convert back, and write with AVAssetWriter. Conversion via Metal is critical — on CPU it takes 15–20 ms per frame just for color space.

For denoising we use Real-ESRGAN or temporal models. Temporal flickering is removed by post-processing: averaging activations from neighboring frames with a weight of 0.1–0.2.

Android: Decoding via MediaCodec into a Surface (OpenGL texture), processing with TFLite GPU Delegate (works directly with textures via setExternalContext()). For Full HD, 256×256 tiling yields ~48 tiles; at 15 ms/tile inference — 720 ms per frame. For quick-enhance, a lightweight model without tiling, downscale to 540p.

Why Temporal Consistency Matters

Independent processing of each frame leads to flickering: details appear and disappear at boundaries. We solve this in two ways: using temporal models with a 3–5 frame window or adding post-processing with activation averaging.

The result is natural-looking video without flickering. Temporal consistency is the key differentiator of a professional solution from simple upscaling.

Real-time and Audio

For real-time we work at reduced resolution: on iPhone 14 Pro — ESRGAN x2 at 480p (~28 ms via ANE), on Snapdragon 8 Gen 2 — via GPU Delegate. We use CameraX with ImageAnalysis and KEEP_ONLY_LATEST strategy.

Audio is copied unchanged: via AVAssetReaderTrackOutput or MediaExtractor, PTS synchronization one-to-one. A common mistake is forgetting to synchronize PTS, which leads to audio drift.

Comparison of Approaches

What's Included in the Work

Timeline Estimates

Post-processing on a single platform with a per-frame model — 3–5 weeks. Both platforms with temporal consistency and real-time mode — 8–14 weeks. Typical post-processing pipeline costs between $8,000 and $15,000.

With 10+ years of experience, over 50 successful projects, and 5 years on the market, we guarantee quality. Get a consultation for your project — we'll analyze your scenarios, select the optimal architecture, and provide accurate timelines. We work turnkey with a result guarantee.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.